Configuring the Adaptive Decision Manager service for on-premises environments

Enable the prediction of customer behavior by configuring the Adaptive Decision Manager (ADM) service. The ADM service creates adaptive models and updates them in real time based on incoming customer responses to your offers. With adaptive models, you can ensure that your next-best-action decisions are always relevant and based on the latest customer behavior.

Note: This procedure is applicable to on-premises environments.
Before you begin: Add at least one ADM node to start the ADM service. For more information, see Adding nodes to Decision Management services.
  1. In the header of Dev Studio, click Configure > Decisioning > Infrastructure > Services > Adaptive Decision Manager.
  2. In the Adaptive decision manager nodes section, click Edit settings.
  3. In the Edit adaptive decision manager settings dialog box, in the Data mart configuration section, specify how often you want to store the model data:
    • To save the data according to the snapshot agent configuration, click Store using agent schedule.
      Note: You can capture the ADM historical data for reporting purposes by using the ADM data mart. The ADM data mart is populated by periodically triggering the ADMSnapshot agent that runs the pzGetAllModelDetails activity that captures the state of models, predictors, and predictor binning. This activity writes that information to a table using the Data-Decision-ADM-ModelSnapshot and Data-Decision-ADM-PredictiveBinningSnapshot classes.
    • To save all the model data every time that you make an update, click Append full historical record at every update, and then go to step 7.
  4. In the Store adaptive model data section, specify what adaptive model data you want to save:
    • To keep all snapshots of ADM models, select Keep all model data.

      This option is required if you use model snapshots for time-based reporting (for example, trend detection). If this option is enabled over a prolonged time period, it can cause database space issues.

    • To keep only the most recent snapshot of ADM models, select Keep only the most recent model data.

      For example, you can use this option if you want to run a report definition of only the most current version of an ADM model.

  5. In the Store predictor bin data section, configure the predictor binning storage preferences:
    • To keep only the most recent snapshot of the predictor binning state, select Keep only the most recent predictor data.

      This option is useful if you want to analyze only the most current status of predictor binning (for example, by using a report definition).

    • To disable capturing the state of predictor binning, select Do not store predictor data.
    • To keep all snapshots of the predictor binning state, select Keep all predictor data.

      Use this option to analyze the changes of predictor binning over time. If this option is enabled over a prolonged time period, the increased number of predictor binning snapshots can cause database space issues because you can have multiple predictors per adaptive model.

    Note: The first time a snapshot is taken, the ADM data mart checks for models that were created prior to Pega 7.2 and migrates them to the current version. After a successful migration, the admmart/modelIdMigrationNeeded dynamic system setting is created and the value is set to false. You do not need to repeat the migration process for successfully migrated instances. To have ADM trigger this step for models that were not migrated yet, change the dynamic system setting to true.
  6. Click Edit agent schedule, modify the default agent schedule configuration of the Pega-DecisionEngine agents, and then click Save.
    For more information about configuring the agent schedule, see Completing the schedule tab.
    For more information about the agents, see Pega-DecisionEngine agents.
  7. In the Service configuration section, define the ADM service parameters:
    1. Select the frequency for checking for model updates.
      Pega Platform keeps a local cache of scoring models. The model update frequency is implemented by periodically triggering the system pulse to retrieve model updates. By default, model updates retrieve the scoring models that are required for running the strategy and the models that are different from those in the local cache.
    2. In the Thread count field, enter the number of threads on all nodes that are running the ADM service.
      The default thread count is the number of available co-processes on that node, minus one.
    3. In the Memory alert threshold field, enter the threshold for triggering the out-of-memory error.
  8. Click Submit.